EXPOSURE PROBLEM IN MULTI-UNIT AUCTIONS

Similar documents
TitleExposure Problem in Multi-Unit Auct.

Supplemental Material: Buyer-Optimal Learning and Monopoly Pricing

Information and uncertainty in a queueing system

Non-Exclusive Competition and the Debt Structure of Small Firms

Professor Huihua NIE, PhD School of Economics, Renmin University of China HOLD-UP, PROPERTY RIGHTS AND REPUTATION

Buyer-Optimal Learning and Monopoly Pricing

Optimal Auctions. Game Theory Course: Jackson, Leyton-Brown & Shoham

Oliver Hinz. Il-Horn Hann

C (1,1) (1,2) (2,1) (2,2)

Asymmetric Information

Auctions. Agenda. Definition. Syllabus: Mansfield, chapter 15 Jehle, chapter 9

Analysis on Mergers and Acquisitions (M&A) Game Theory of Petroleum Group Corporation

Game Theory. Lecture Notes By Y. Narahari. Department of Computer Science and Automation Indian Institute of Science Bangalore, India July 2012

Mechanism Design and Auctions

Multiunit Auctions: Package Bidding October 24, Multiunit Auctions: Package Bidding

Auctions That Implement Efficient Investments

Individual Comparative Advantage and Human Capital Investment under Uncertainty

1 < = α σ +σ < 0. Using the parameters and h = 1/365 this is N ( ) = If we use h = 1/252, the value would be N ( ) =

Non-Inferiority Tests for the Ratio of Two Correlated Proportions

Multiple-Project Financing with Informed Trading

Auction Theory: Some Basics

Chapter 3. Dynamic discrete games and auctions: an introduction

LECTURE NOTES ON MICROECONOMICS

An Equilibrium Analysis of the Simultaneous Ascending Auction

Lecture 6 Applications of Static Games of Incomplete Information

INDEX NUMBERS. Introduction

Mechanism Design and Auctions

Forward Vertical Integration: The Fixed-Proportion Case Revisited. Abstract

Quantitative Aggregate Effects of Asymmetric Information

Auctions. Michal Jakob Agent Technology Center, Dept. of Computer Science and Engineering, FEE, Czech Technical University

Games with more than 1 round

Game Theory Lecture #16

Effects of Size and Allocation Method on Stock Portfolio Performance: A Simulation Study

Withdrawal History, Private Information, and Bank Runs

Auction is a commonly used way of allocating indivisible

Economic Performance, Wealth Distribution and Credit Restrictions under variable investment: The open economy

BIS Working Papers. Liquidity risk in markets with trading frictions: What can swing pricing achieve? No 663. Monetary and Economic Department

Auctions. Michal Jakob Agent Technology Center, Dept. of Computer Science and Engineering, FEE, Czech Technical University

NBER WORKING PAPER SERIES SELF-FULFILLING CURRENCY CRISES: THE ROLE OF INTEREST RATES. Christian Hellwig Arijit Mukherji Aleh Tsyvinski

On the Impossibility of Core-Selecting Auctions

Inventory Systems with Stochastic Demand and Supply: Properties and Approximations

VI Introduction to Trade under Imperfect Competition

ECONOMIC GROWTH CENTER

Optimal Mixed Spectrum Auction

Management Accounting of Production Overheads by Groups of Equipment

ECON Microeconomics II IRYNA DUDNYK. Auctions.

Does Hedging Reduce the Cost of Delegation?

Quality Regulation without Regulating Quality

Physical and Financial Virtual Power Plants

Corporate Finance: Credit rationing. Yossi Spiegel Recanati School of Business

Interest Rates in Trade Credit Markets

A NOTE ON SKEW-NORMAL DISTRIBUTION APPROXIMATION TO THE NEGATIVE BINOMAL DISTRIBUTION

We connect the mix-flexibility and dual-sourcing literatures by studying unreliable supply chains that produce

Cash-in-the-market pricing or cash hoarding: how banks choose liquidity

No. 81 PETER TUCHYŇA AND MARTIN GREGOR. Centralization Trade-off with Non-Uniform Taxes

Asian Economic and Financial Review A MODEL FOR ESTIMATING THE DISTRIBUTION OF FUTURE POPULATION. Ben David Nissim.

Recap First-Price Revenue Equivalence Optimal Auctions. Auction Theory II. Lecture 19. Auction Theory II Lecture 19, Slide 1

An Ascending Double Auction

Advertising Strategies for a Duopoly Model with Duo-markets and a budget constraint

1 Theory of Auctions. 1.1 Independent Private Value Auctions

Notes on Auctions. Theorem 1 In a second price sealed bid auction bidding your valuation is always a weakly dominant strategy.

ECON 400 Homework Assignment 2 Answer Key. The Hicksian demand is the solution to the cost minimization problem.

March 30, Why do economists (and increasingly, engineers and computer scientists) study auctions?

Too much or not enough crimes? On the ambiguous effects of repression

U. Carlos III de Madrid CEMFI. Meeting of the BIS Network on Banking and Asset Management Basel, 9 September 2014

Core Deviation Minimizing Auctions

Informal Lending and Entrepreneurship

Microeconomic Theory II Preliminary Examination Solutions Exam date: June 5, 2017

CS364B: Frontiers in Mechanism Design Lecture #18: Multi-Parameter Revenue-Maximization

: now we have a family of utility functions for wealth increments z indexed by initial wealth w.

October An Equilibrium of the First Price Sealed Bid Auction for an Arbitrary Distribution.

Volumetric Hedging in Electricity Procurement

The Impact of E-Commerce on Competition in the Retail Brokerage Industry

In ation and Welfare with Search and Price Dispersion

Gathering Information before Signing a Contract: a New Perspective

Parkes Auction Theory 1. Auction Theory. Jacomo Corbo. School of Engineering and Applied Science, Harvard University

Adverse Selection in an Efficiency Wage Model with Heterogeneous Agents

Diversification: more than one project. It may be beneficial for a firm, in terms of getting hold of external funds, to have several projects.

The Strategic Effects of Parallel Trade ~Market stealing and wage cutting~

Mechanism Design and Auctions

A Note on Reliefs for Traveling Expenses to Work

Bayesian games and their use in auctions. Vincent Conitzer

CS 573: Algorithmic Game Theory Lecture date: 22 February Combinatorial Auctions 1. 2 The Vickrey-Clarke-Groves (VCG) Mechanism 3

Side-Communication Yields Efficiency of Ascending Auctions: The Two-Items Case

Working Paper Series. This paper can be downloaded without charge from:

A Comparative Study of Various Loss Functions in the Economic Tolerance Design

Gottfried Haberler s Principle of Comparative Advantage

KIER DISCUSSION PAPER SERIES

Byungwan Koh. College of Business, Hankuk University of Foreign Studies, 107 Imun-ro, Dongdaemun-gu, Seoul KOREA

Matching Markets and Social Networks

Independent Private Value Auctions

Recalling that private values are a special case of the Milgrom-Weber setup, we ve now found that

A Multi-Objective Approach to Portfolio Optimization

The Optimal Entry Mode of the Multinational Firm under Network Externalities: Foreign Direct Investment and Tariff

Fear of Losing in Dynamic Auctions: An Experimental Study. Peter Cramton, Emel Filiz-Ozbay, Erkut Y. Ozbay, and Pacharasut Sujarittanonta 1

HW Consider the following game:

ECON20710 Lecture Auction as a Bayesian Game

Auctions Introduction

***SECTION 7.1*** Discrete and Continuous Random Variables

All Equilibrium Revenues in Buy Price Auctions

Transcription:

EXPOSURE PROBLEM IN MULTI-UNIT AUCTIONS Hikmet Gunay and Xin Meng University of Manitoba and SWUFE-RIEM January 19, 2012 Abstract We characterize the otimal bidding strategies of local and global bidders for two heterogenous licenses in a multi-unit simultaneous ascending auction. The global bidder wants to win both licenses to enjoy synergies; therefore, she bids more than her stand-alone valuation of a license. This exoses her to the risk of losing money even when she wins all licenses. We determine the otimal bidding strategies in the resence of an exosure roblem. By using simulation methods, first, we show the frequency of inefficient allocation in the simultaneous ascending auction. Then, we show that the Vickrey-Clarke-Groves (VCG) mechanism may generate more revenue than the simultaneous ascending auction. JEL Codes:D44, D82 Keywords: Multi-Unit Auctions, Vickrey Clarke Groves (VCG) mechanism, Exosure Problem, Synergies, Comlementarity PRELIMINARY DRAFT. Acknowledgements will be added later. 1

1 Introduction In a tyical American or Canadian sectrum license auction, hundreds of (heterogenous) licenses are sold simultaneously. Each of these licences gives the sectrum usage right of a geograhical area to the winning firm. Some local firms are interested in winning only secific licenses in order to serve local markets while other global firms are interested in winning all the licenses in order to serve nationwide. 1 The global firms enjoy synergies if they win all the licenses which gives them an incentive to bid over their stand-alone valuations for some licenses. As a result, there is a risk of incurring losses. Therefore, global bidders lower their bids. This is known as the exosure roblem. 2 In a model simlifying the American and the recent Canadian sectrum license auctions, we derive the otimal bidding strategies of local and global firms in a simultaneous ascending auction of two licenses. We mainly focus on the otimal bidding strategies when there is the ossibility of an exosure roblem, and through simulations, we determine how frequently the exosure roblem (i.e., ex-ost loss) occurs. In addition, we decomose the frequency into two cases; the case in which the exosure roblem occurs when the global bidder wins only one license, and the case in which the exosure roblem occurs when the global bidder wins all licenses. Exosure roblem indicates that the allocation may not be efficient. We comare the efficiency and revenue roerties of the simultaneous ascending auction with those of the Vickrey-Clarke-Groves (VCG) mechanism when bidders are allowed to bid on ackages. VCG is an efficient auction that gives the highest revenue among all incentive comatible, individually rational, efficient auctions. In the literature (e.g., Ausubel and Milgrom (2006)), there are examles which show that VCG mechanism may give extreme low revenue in comlete information settings. We show that VCG mechanism may give higher revenue to 1 In the recent Canadian Advanced Wireless Sectrum auction, firms such as Globalive and Rogers were interested in all licenses whereas firms such as Bragg Communication and Manitoba Telecom Services (MTS) were interested in East Coast and Manitoba licenses, resectively. 2 We will interchangeably use exosure roblem as follows. We say that an exosure roblem occurred whenever the global bidder incurs a loss ex-ost. 2

the seller for many arameter saces and various distributions in incomlete information setting. We also show the frequency of inefficient allocation when simultaneous ascending auction is used. The multi-unit auction literature generally assumes that global bidders have either equal valuations (Englmaier et. al (2009), Albano et. al. (2001), Kagel and Levin (2005), Katok and Roth (2004), Rosenthal and Wang (1996), and Krishna and Rosenthal (1996)) or very large synergies (Albano et al. (2006)). The sectrum licenses for different geograhic areas are not homogenous objects; hence, the equal valuation assumtion does not fit the Canadian or the American sectrum license auction. Moreover, in a heterogeneous license environment, bidders may not dro out of both auctions simultaneously. This enables us to analyze bidding behavior in the remaining auction, and hence, the exosure roblem in detail. We allow for moderate synergies, and our focus is on the exosure roblem and the comarison of revenue and efficiency roerties of the simultaneous ascending auction with those of the VCG auction, unlike Albano et.al (2006). 3 In our aer, the global bidder will lower his bid because of the exosure roblem; however, their otimal strategy still requires him to bid over his stand alone valuation for at least one license. If he wins this license by receiving a otential loss, then he may need to stay in the other license auction to minimize his loss. Therefore, there are cases in which the exosure roblem may arise even when the bidder wins all the licenses. Two additional aers related to this aer are Goerre and Lien (2010) and Zheng (2008). Goerre and Lien assume that the valuation of winning a given number of licenses is the same regardless of the comosition. Hence, they find that the otimal dro out rice is the same for both licenses. In our aer, marginal valuations are different. Global bidder s valuation of license A or license B is different. 4 Hence, our aer shows that the otimal dro out rice 3 They assume large synergies so no exosure roblem exists in equilibrium. Our results coincide with theirs when we also assume large synergies. 4 We model situations in which winning the sectrum license for, say, Iowa City is different than winning the license for New York city. This flexibility comes at the exense of allowing only one global bidder. Goerre and Lin (2010) s assumtions allow them to write a tractable model in which they can allow multile global bidders. 3

is not the same, and the exosure roblem of this case will lead to different results such as winning all licenses and still incurring a loss. In addition, through simulations, we show the frequency of inefficient allocation for the simultaneous ascending auction. Our common oint with Goerre and Lien (2010) is that we also find VCG mechanism may give higher revenue than the simultaneous ascending auction. Zheng (2008) is mainly interested in showing that jum-bidding will alleviate the exosure roblem. We do not allow jum-bidding as the Canadian sectrum auction has not allowed this. Our aer can be contrasted with Kagel and Levin (2005) and Krishna and Rosenthal (1996). Krishna and Rosenthal (1996) study a second rice auction (simultaneous and sequential auctions), and do not secifically analyze exosure roblem. Kagel and Levin (2005) use a single global bidder that cometes with several local bidders. Their ascending bid version of the uniform rice auction is different than ours since their two goods are sold in a single auction. In the Canadian sectrum auction, the licenses are sold in searate auctions so we use different auctions in our model. We also use simulations to calculate the robability of exosure roblem occurrence, and the comarison of the revenue and efficiency roerties of this auction with the VCG auction. One of the contributions of this aer is to use the simulation methods to show the robability of the exosure roblem occurring in the simultaneous ascending auction. We also decomosed this robability into two cases; when the global bidder wins only one license, and when he wins both licenses. We call the first one exosure roblem I and the latter one exosure roblem II. We use four different distributions to draw the valuations; uniform, beta distribution with alha and beta equal to two, beta distribution with alha equal to one and beta equal to four, and beta distribution with alha equal to four and beta equal to one. We comare our the revenue and the efficiency roerties of this auction with those of the VCG auction. When we use uniform distribution and one local bidder on each licences, we show that the revenue is 8 er cent higher but the allocation is inefficient 4 er cent of the time. 4

Almost all roofs are included in the Aendix. 2 The Model There are 2 licenses, license A and B for sale. 5 There are one global bidder who demands both licenses and m j = m 1 local bidders who demand only license j = A, B. Secifically, m j will denote the number of active local bidders on the auction. 6 Both local bidders and the global bidder have a rivate stand alone valuation for a single license, v ij, where i and j reresent the bidder and the license, resectively. The valuations v ij are drawn from the continuous distribution function F (v ij ) with suort on [0, 1] and robability density function f(v ij ) which is ositive everywhere with the only excetion that f(0) 0 is allowed. The bidders tye, global or local, is ublicly known. We consider a setting where the licenses are auctioned off simultaneously through an ascending multi-unit auction. Each license is auctioned off at a different auction (like Krishna and Rosenthal (1996) but unlike Kagel and Levin (2005)) but at the same time. Prices start from zero for both licenses and increase simultaneously and continuously at the same rate. Bidders choose when to dro out. When only one bidder is left on a given license, the clock stos for that license, and the sole remaining bidder wins the license at the rice at which the last bidder droed out. If there are more than one bidder remaining on the other license, its rice will continue to increase. If n bidders dro out at the same rice and nobody is left in the auction, then each one of them will win the license with robability 1 n. The dro-out decision is irreversible. Once a bidder dros out of bidding for a given license, he cannot bid for this license later. 7 The number of active bidders and the droout rices are ublicly known. We also assume that there is no budget constraints for the bidders. 5 We use two licenses like Albano et. al. (2001 and 2006), Brusco and Loomo (2002), Chow and Yavas (2009), and Menucicci (2003). 6 Allowing different number of local bidders er license will not change our qualitative results. 7 In the real-world auctions, there is activity rule. If the bidders do not have enough highest standing bids, then the number of licenses they may bid on is decreased (in the next rounds). Hence, when there are two licenses, this translates into an irreversible dro-out. 5

We assume that there is a homogeneous ositive synergy for the global bidder. Secifically, letting bidder 1 be the global bidder, the global bidder s total valuation, given that it wins two licenses is, V 1 = + v 1B + α, where the synergy term α is assumed to be strictly ositive and ublic knowledge. 8 His stand-alone valuation of license A or B is given by or v 1B. Bidder ia, i = 2, 3,...m is only interested in license A, and her rivate valuation is v ia. Bidder ib is only interested in license B, and her rivate valuation is v ib. 9 A local bidder who is interested in license j articiates only in license j auction. We derive a symmetric erfect Bayesian equilibrium through a series of lemmas that follow. First, we describe the equilibrium strategy of the local bidder. Lemma 1 Each local bidder has a weakly dominant strategy to stay in the auction until the rice reaches his stand alone valuation. This is a well-known result so we ski the roof. Now, suose all the local bidder dros out of license B auction, and hence, the global bidder wins license B at the rice B, which in equilibrium is equal to B = max{v 2B,..., v (m)b } by lemma 1. Then, as the rice for license A increases, the global bidder will comare the ayoff from droing out from license A auction at the clock rice (which is v 1B B ) and the ayoff from winning license A at rice (which is + v 1B + α B ). The udated otimal dro out rice, A, is found by equating these two equations: +v 1B +α B A = v 1B B A = + α. If global bidder wins license A first, the udated otimal dro out rice, B, can be found symmetrically. We state this as lemma 2. Lemma 2 If the global bidder wins license B (or A) first, then it will stay in license A (or B) auction until the rice reaches + α (or v 1B + α) The global bidder will not dro out before the rice reaches his minimum of stand-alone valuations. Otherwise, they will lose the chance of winning both licenses and enjoying the 8 Public knowledge assumtion can be removed, and all results are still valid. We assume ublic knowledge not to comlicate the notation. 9 We do not assume that v ia > v ib since local firms are different; hence, their efficiency may differ. 6

synergy. In addition, if the global bidder s average valuation, V 1 2 = +v 1B +α 2, exceeds 1, bidding u to his average valuation will shut out the local bidders since local bidders stand alone valuation can be at most 1.If α is large enough, this condition will always be satisfied. In such a case, the global bidder always wins both licenses in equilibrium. We summarize these results as lemma 3. Lemma 3 a) The global bidder stays in both license auctions at least until the rice reaches the minimum of his stand-alone valuations. b) If his average valuation is greater than 1, the global bidder s equilibrium strategy is to stay in until the rice reaches his average valuation. To calculate the otimal dro out rice for the global bidder, consider first the case in which > v 1B. 10 The global bidder must comare the ayoffs for two cases at each rice as the clock is running: Case 1 is the ayoff from droing out from license B auction at rice and otimally continuing on license A auction. Case 2 is the ayoff from winning license B at rice and otimally continuing on license A auction. 11 At the beginning of the auction, that is = 0, the second case ayoff is higher so the global bidder will start by staying in the auction. We show that the difference between these two cases are monotonic in ; therefore, there is a unique rice that makes the global bidder indifferent between these two cases (assuming that the local bidders are still active). This is the otimal dro out rice, 1. We show that this rice can be calculated at the beginning of the auction. Note that according to Lemma 3, 1 v 1B, and the otimal udated dro out rice for license A, after winning license B at rice, is + α. We denote the exected rofit of the global bidder for Case 1 by EΠ 1 1 and his exected rofit for Case 2 by EΠ 2 1, resectively. Let A = max{v 2A,..., v (m+1)a } be the rice the global bidder will ay for the license A, if he wins license A. Payoffs are as follows: 10 The other case can be calculated symmetrically. 11 The global bidder will dro out of license B first since > v 1B, if he has not won license A yet. 7

EΠ 1 1 = Max{0, v1a ( A )g( A )d A } (1) EΠ 2 1 = Min{v1A +α,1} (V 1 A )g( A )d A + 1 Min{ +α,1} (v 1B )g( A )d A (2) The exlanation of equation 1 is as follows. After the global bidder dros out of the auction for license B at, it becomes just like a local bidder, and hence, will continue to stay in the auction for license A until. If he wins, he will ay A since the local bidder with highest valuation of license A will dro out last (by Lemma 1). In order to calculate his exected rofit, global bidder will be using G( A ) (highest order statistic) which is the distribution function of the local bidders highest valuation A for license A given. When there are m 1 local bidders in license A, the distribution function G( A ) and its density function g( A ) are: A f(v)dv G( A ) = (F ( A )) m 1 = ( 1 )m 1 (3) f(v)dv A f(v)dv f( A ) g( A ) = (m 1)( 1 f(v)dv )m 2 ( 1 ). (4) f(v)dv The first term of EΠ 2 1 is Firm 1 s exected rofit of winning both licenses; assuming that he wins license B at the rice. If the highest local bidder s valuation A is less than the global bidder s (udated) willingness to ay, + α, then the global bidder wins license A and ays A. Since A < 1, we use the minimum function in the uer limit of the first integral. The second term of EΠ 2 1 is Firm 1 s exected rofit of winning only license B which can haen only if A > + α. Note that the second term is non-ositive by Lemma 3 (which is the exosure roblem arising from winning only one license). In Lemma 4 below, we characterize the global bidder s equilibrium bids. It can be found from EΠ 1 1 = EΠ 2 1. Note that these ayoffs are changing as local bidders bidding for A are droing out; that is, m 1 is changing. Therefore, the lemma below gives the global 8

bidder s (udated) equilibrium dro out rice as the local bidders change. We show, in the roof, that this udated rice increases as local bidders dro out. Lemma 4 : Suose that the average valuation of the global bidder is less than 1 and no local bidders have droed out yet. If > v 1B, the global bidder 12 will dro out of license B auction at the unique otimal dro-out rice 1 [0, 1] that satisfies EΠ 1 1 = EΠ 2 1. Moreover, a) If + α < 1, and +α G( A )d A + (v 1B ) < 0, then 1 < and the global bidder will stay in license A auction until (after droing out from license B auction). b) If + α < 1, and +α G( A )d A + (v 1B ) > 0, then 1 > and the global bidder will also dro out of license A auction at 1. c) If + α > 1, and 1 G( A )d A + (v 1B + α 1) < 0, then 1 < and the global bidder will stay in license A auction until (after droing out from license B auction). d) If + α > 1, and 1 G( A )d A + (v 1B + α 1) > 0, then 1 > and the global bidder will also dro out of license A auction at 1. Proof. See the Aendix. We are ready to summarize our Perfect Bayesian equilibrium. Proosition 5 (Perfect Bayesian Equilibrium) a) Local bidder j = {A, B} l = {2, 3,..m 1} of each license will stay in the auction j until rice reaches their valuation v kj. b) A global bidder active only on license j will bid v 1j + α, if he won license k = j. He will bid v 1j when he did not win license k. c) When > v 1B and the average valuation is less than one, the global bidder who is active on both licenses and facing m 1 active local bidders on license A will dro out from license B at the rice that equates equations 1 and 2. d) When < v 1B and the average valuation is less than one, the global bidder who is active on both licenses and facing m 1 active local bidders on license B will dro out from license A at the rice that equates equations 1 and 2 (symmetrically relaced with v 1B. 12 If < v 1B, then the roosition has to be written symmetrically 9

e) If the average valuation is greater than one, the global bidder will stay in both auctions until rice reaches his average valuation. f) Out-of-equilibrium-ath beliefs: When a bidder dros out, the other bidders will see this as an equilibrium behavior. Hence, there is no need to secify the out of equilibrium ath beliefs. At the beginning of the game, each bidder calculates its otimal dro-out rice. For local bidders, the otimal dro out rices are their valuations. In equilibrium, it is otimal for the global bidder to stay in the auctions for both licenses u to his otimal dro-out rice calculated in Lemma 4. When his average valuation exceeds 1, he will stay until this average valuation and win both licenses. When the rice reaches the minimum of these otimal dro-out rices, that bidder dros out of license auction. If, for examle, the highest local bidder for license B droed out before the global bidder, the global bidder would continue to stay in the auction for license A until the rice reaches +α. At this rice, he finds that the ayoff from winning only license B is more than the ayoff from winning both licenses even though it will enjoy synergy; hence, it dros out. If the value of the licenses were identical (e.g. Albano et. al. (2001)), the global firm would dro out of both licenses at the same time. In this case, our Lemma 4 art b will be valid; that is, 1 > = v 1B, hence, the global bidder dros out from both licenses at the same time. This result coincides with Albano et. al. (2001),(2006) and Goerre and Yuanchuan (2010). The following is a corollary of Lemma 4, and is an examle for the otimal dro out rice when F (.) is a uniform distribution. Corollary 6 : Assume that valuations are drawn from a uniform distribution with a suort [0, 1]. In addition, assume that > v 1B (other case is symmetrically found by exchanging 10

When A is here, global bidder wins both but makes a loss When A is here, global bidder wins B and makes a loss Exosure Problem II Exosure Problem I 0 v 1B B 1 + v 1B + α B + α 1 Figure 1: EXPOSURE PROBLEM with v 1B ), and there is one local bidder in each license. 1 = 1 2 {v 1B + α + 1 (v 2 1B + 1 2v 1B α 2 + 2v 1B α + 2α 4 α) 1 2 }, if 0 < < 1 α and 2(1 )( v 1B ) > α 2 ; 1 3 { + v 1B + α + 1 (( + v 1B + α + 1) 2 3( + α) 2 6v 1B ) 1 2 }, if 0 < < 1 α and 2(1 )( v 1B ) α 2 ; 1 2 {v 1B + α + 1 {(v 1B + α + 1) 2 4( + v 1B + α) + 2 + 2v 2 1A } 1 2 }, if 1 α < 1 and 1 + > 2(v 1B + α); 2( +v 1B +α) 1 3, if 1 α < 1 and 1 + 2(v 1B + α). (5) The otimal dro-out rice is a function that takes a unique value defined in the corollary above. For examle, case 0 < < 1 α and 2(1 )( v 1B ) > α 2 imlies that 1 <. 2.1 Exosure Problem We now can discuss the exosure roblem with the hel of Figure 1. In the first tye of exosure roblem, the global bidder may win license B at a rice above his stand alone valuation (i.e., v 1B < B < 1) and lose the other license (i.e., A > + α. This is the tye of exosure roblem Chakraborty (2004) focuses on. In the second tye of exosure roblem, the global bidder wins both licenses but incurs a loss. This is the case when he wins license B at v 1B < B < 1 and wins license A at + α > A > + α + v 1B B. Note that if he wins license A at the rice + α + v 1B B, his ayoff is zero. The global bidder stays in the auction for license A in order to minimize its loss from winning only license B even if the rice asses + α + v 1B B. If objects were homogenous, the second tye of exosure roblem would not be observed since the bidder would dro out of both license auctions at the same time. 11

Table 1: PROBABILITY OF EXPOSURE PROBLEM Percentage of Percentage of Total Percentage of Exosure Problem 1 Exosure Problem 2 Percentage Inefficiency One One One One α Local Bidder Local Bidder Local Bidder Local Bidder Beta Distribution with arameters α = 1 and β = 4 0.2 2.85 0.64 3.48 7.84 0.4 2.29 2.43 4.72 5.45 0.6 0.54 1.59 2.13 2.15 0.8 0.04 0.62 0.66 0.66 Uniform Distribution 0.2 1.17 0.40 1.57 3.29 0.4 1.13 1.07 2.20 4.22 0.6 0.69 1.51 2.20 4.01 0.8 0.36 1.25 1.61 2.62 Beta Distribution with arameters α = 2 and β = 2 0.2 0.86 0.29 1.16 5.10 0.4 0.80 1.31 2.12 4.75 0.6 0.82 2.31 3.14 3.88 0.8 0.22 1.65 1.86 1.92 Beta Distribution with arameters α = 4 and β = 1. 0.2 0.34 0.77 1.11 2.72 0.4 0.20 0.93 1.13 1.95 0.6 0.00 0.50 0.50 0.64 0.8 0.00 0.26 0.26 0.28 2.2 Simulations In the following, through simulations, we determine the robability of the occurrence of exosure roblems under various environments. As we noted, we say that exosure roblem occur when the global bidder wins one or both licenses with a loss ex-ost. We have used MATLAB to write our simulation code. This code first draws the valuations for both the global and the local bidders from a given distribution function. One set of valuations corresond to one auction. We, then, calculate the otimal dro-out rice of the global bidder; local bidders dro-out rices are their valuations. If some local bidders valuations are lower than the global bidder, the global bidder s dro-out rice is udated as these local bidders 12

dro out from the auction. If the global bidder does not win the first license, then no exosure roblem occurs. If he wins the first license above its valuation for that license, then we calculate his udated rice for the remaining license (unless the global bidder dros out from both licenses at the same time). We next determine whether the global bidder will win the remaining license at a ositive rofit (no exosure roblem), at a loss (exosure roblem II) or lose the remaining license (exosure roblem I). Dividing the number of each of these events to the number of draws yields the robability of each event. 13 We use four different distribution functions to draw valuations: uniform, beta distribution with α = β = 2, beta distribution with α = 1 and β = 4, and beta distribution with α = 4 and β = 1. The second distribution is a mean reserving sread of the uniform distribution. The third one is first order stochastically dominated by the uniform distribution, while the fourth one first order stochastically dominates the uniform distribution. We use one local bidder on each license. 14 We run simulations for four different synergy levels: 0.2, 0.4, 0.6, 0.8. 15 The 0.2 reresents for small synergy, 0.4 and 0.6 reresents middle synergy, and 0.8 reresents a large synergy level. We reort the results in Table 1. We find that the global bidder may face exosure roblem with robability 4.33 er cent, if the valuations are drawn from beta distribution with arameters α = 1 and β = 4, and the synergy level is equal to 0.4. Table 1 also shows that the exosure roblem occurs with the smallest robability among all these different distributions when the synergy level is 0.8. This is exected since the global bidder can bid very high for the remaining license (in most cases more than 1) but not face exosure roblem much. 13 In our simulation, to simlify calculations, we only consider cases in which the global bidder values the license B more than license A. After 15000 draws, we select only the valuations where the global bidder s valuation for license A is greater than license B. Hence, we are left with aroximately 7500 draws. We used UNIX system of the University of Manitoba, and our latos for the simulations. In the UNIX machine, it took more than four days to run each code. 14 Given the comlexity of the code we use, we feel that using one local bidder in our simulations are enough to draw reasonable conclusions though this can be extended to two local bidders in each license. Using three local bidders or more would extremely comlicate the code since one has to kee track of udated rices every time a local bidder dros off. 15 We also write codes for a finer synergy level of 0.1,0.2,...0.9 but we do not reort them since no new insight is learned from this. 13

In our simulations, Exosure roblem I occurred most often when the synergy level is 0.4. In this case, the global bidder overbids to enjoy the middle level synergy, so he is very likely to make a (substantial) loss when he wins the first license. His otimal dro-out rice for the remaining license is generally below 1; hence, the risk of losing the second license is high, and this is the reason for observing exosure roblem I for the synergy level of 0.4. Exosure roblem II generally occurs the most when the synergy level is 0.6. After winning the first license, the global bidder will stay in the remaining license auction for a higher dro out rice; hence, rather than exosure roblem I, exosure roblem II is likely to occur. Of the uniform and beta distribution with arameters α = β = 2, we see that exosure roblem is more likely to occur in the latter one. This is exected since the valuations are more likely to be on the extreme sides. Hence, local bidders are more likely to have high valuations for the remaining license, and force the global bidder to dro out or make a loss. Of the two beta distribution, the one with α = 4 and the one with α = 1, we observe that the exosure roblem occurs much less frequently with the first one. The reason is that with the former one, the valuations of the global bidder is likely to be higher; hence, even with a small synergy level, their otimal udated dro out rice after winning one license is more than one in most cases. Therefore, it is less likely to have exosure roblem. On the other hand, when the global bidder s udated otimal rice is less than one, it is also more likely that the local bidder s valuation will be high so in such cases, one may see exosure roblem. As the synergy increases, this case is less likely to haen. In Figure 2.2, we show that, for uniform distribution, the allocation is inefficient 8 er cent of the time when α = 0.2. As α increases, global bidder wins both licenses more often without facing exosure roblem that much, and this is the efficient outcome. Hence, we observe inefficient allocation only 3 er cent of the time for α = 0.8. Most of the inefficiency is due to exosure roblem. 14

0.045 0.04 0.035 Plot of Probability of Inefficiency Probability of Inefficiency Total Probability of Exosure Problem Probability of Exosure Problem 1 Probability of Exosure Problem 2 0.03 robability 0.025 0.02 0.015 0.01 0.005 0 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 alha (Synergy) 3 Comarison with Vickrey Clarke Groves Auction In this section, we will comare the revenue of our auction with the revenue of Vickrey Clarke Groves (VCG) auction. VCG auction maximizes the exected ayment of each agent among all mechanisms for allocating multile objects that are efficient, incentive comatible, and individually rational. 16 In this auction, the seller will let the bidders bid on license A, license B and the whole ackage license A and B. We, will first calculate the ayment of each winner for all cases; that is, calculate the revenue of the seller. The ayment of a winner (say layer i) in this auction is the difference between the social welfare of the others if the bidder did not articiate in the auction (denote this as W i (x i ) where x i denote the bid of all layers other than layer i), and the welfare of the others when he articiated in the auction, and bid truthfully (denote 16 See roosition 16.2 of Krishna (2010) 15

this as W i (x), where x denote the bid of all layers.) since truthful bidding is the weakly dominated strategy. The table below shows the valuations of the bidders. To give an examle of how ayments are calculated, let us assume that v 2A + v 1B > + v 1B + α > v 2A + v 3B. VCG auction will allocate license A to local bidder A, and license B to global bidder. Payment of local bidder A is (W 2 (x 2 )) W 2 (x) = ( + v 1B + α) v 1B = + α. If local bidder A does not articiate in the auction, then global bidder will win the ackage; hence, the welfare of the others W 2 (x 2 ) = + v 1B + α. When it articiates in the auction, global bidder gets only license B, and local bidder B gets nothing; hence, the welfare of others in this case is W 2 (x) = v 1B. A B AB v 1B +v 1B +α v 2A 0 v 2A 0 v 3B v 3B Payment of the global bidder is: (W 1 (x 1 )) W 1 (x) = (v 2A + v 3B ) v 2A = v 3B. If the global bidder does not articiate in the auction, local bidder A and B wins each license;hence, the welfare of others is the term inside the arenthesis. When the global bidder articiates in the auction, local bidder A wins license A but the welfare of local bidder B is zero; hence the welfare of others in this case is just v 2A. The total revenue of the seller in this case will be + α + v 3B. We summarize the revenue of the seller for all cases in the following roosition. 17 Proosition 7 Suose that there is one global bidder and one local bidder bidding for each license. In the VCG auction, the seller s revenue will be as follows deending on the valuations of the bidders. CASE I: Suose that the valuations are such that + v 1B + α < v 2A + v 3B. 17 This result can easily be extended to one global and many local bidders case. Since, we use one global bidder and one local bidder in each license in the simulations, to save the notation, we give the result for a secial case. 16

A) (Local bidders win each license) And suose that < v 2A and v 1B < v 3B. There are four sub cases to consider. i) v 3B > v 1B + α and v 2A > + α, then the revenue is + v 1B. ii) v 3B < v 1B + α and v 2A < + α, then the revenue is 2( + v 1B + α) v 3B v 2A. iii) v 3B > v 1B + α and v 2A < + α, then the revenue is 2 + v 1B + α v 2A. iv) v 3B < v 1B + α and v 2A > + α, then the revenue is + 2v 1B + α v 3B. B) (Local bidder wins A, and global bidder wins B) And suose that v 2A > and v 3B < v 1B. Then, the revenue is + α + v 3B. C) (Local bidder wins B, and global bidder wins A) And suose that v 2A < and v 3B >. Then, the revenue is v 1B + α + v 2A CASE II: Suose that the valuations are such that + v 1B + α > v 2A + v 3B. A) (Global bidder wins both licenses) And suose that < v 2A and v 1B < v 3B. Then, the revenue is v 2A + v 3B. B) (Global bidder wins both licenses) And suose that > v 2A and v 1B > v 3B. Then, the revenue is v 2A + v 3B. C) (Global bidder wins license A, local bidder B wins license B) And suose that v 2A < and v 3B > v 1B + α. Then, the revenue is v 2A + v 1B + α. D) (Global bidder wins license B, local bidder A wins license A) And suose that v 2A > + α and v 3B < v 1B. Then, the revenue is v 3B + + α. We comare the revenue of the simultaneous ascending auction with those of the VCG auction through simulation methods. Our results are summarized in Figure 3. We run the simulations for α = 0.2, 0.4, 0.6, 0.8. When α = 0.2, the revenue is 10 er cent higher in the simultaneous ascending auction. 18. When α = 0.8, global bidder must be winning the auction most of the time, and they ay a total of v 2A + v 3B. The exected ayment would be 1 in the uniform distribution, which we observe in Figure 3. 18 In the 2008 Canadian sectrum auction, the revenue was 4 billion dollars so 10 er cent is a significant number in our view. 17

0.86 0.84 Plot of Revenue Comarison Average Revenue for Our Model Average Revenue for efficient VCG Model 1 0.98 Plot of Revenue Comarison Average Revenue for Our Model Average Revenue for efficient VCG Model 0.82 0.96 Average Revenue 0.8 0.78 0.76 0.74 Average Revenue 0.94 0.92 0.9 0.88 0.72 0.86 0.7 0.84 0.68 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 alha (Synergy) 0.82 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 alha (Synergy) 1.62 Plot of Revenue Comarison Average Revenue for Our Model Average Revenue for efficient VCG Model 0.42 Plot of Revenue Comarison Average Revenue for Our Model Average Revenue for efficient VCG Model 1.6 0.4 Average Revenue 1.58 1.56 1.54 Average Revenue 0.38 0.36 0.34 1.52 0.32 1.5 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 alha (Synergy) 0.3 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 alha (Synergy) Figure 2: Uniform Distribution, To Left; Beta Distribution with α = 2, β = 2, To Right; Beta Distribution with α = 4, β = 1, Bottom Left; Beta Distribution with α = 1, β = 4, Bottom Right 18

4 More Than One Global Bidder In this section, we will analyze the case where there is more than one global bidder. The difficulty in this case arises from the fact that a global bidder s otimal dro out rice deends on the other global bidders otimal dro out rice. Then, to calculate the otimal dro out rices, one should solve more than one non-linear equations simultaneously. To make things worse, the global bidder should know the distribution of the other global bidder s dro out rice when making its own dro out rice calculation. This makes deriving an analytical result imossible. To make our case, assume that there are two global bidders, Firm 1 and Firm 2, and one local bidder on each of license A and license B. Everything else is the same as the revious section. Firm 1 will make a similar calculation as we discussed in one global bidder case excet that the rice would deend on the dro out rice of the second global bidder which we denote as 2 Secifically, he should make the following calculation while deciding to stay in the license B auction or not. This calculation is done based on a history in which none of the other bidders have droed out yet. [ EΠ 1 1 = Max{0, E ( [ v1a v1a Max{v 2A, v 3A }P ( 2 < v 4B )dg(max{v 2A, v 3A } ) + (6) ] Max{(v 2A + α), v 3A }P ( 2 > v 4B )dg(max{(v 2A + α), v 3A } )] 2 } (7) EΠ 2 1 = Min{v1A +α,1} (V 1 A )dg(max{v 2A, v 3A } )+ 1 Min{ +α,1} (v 1B )dg(max{v 2A, v 3A } ) The exlanation of equation 6 is as follows. If this global bidder dros out from license B before the other global and local bidders, then it will continue on license A auction as a local bidder. Hence, it can only derive a benefit of if it wins the license. The rice it will ay for license A deends on the result of whether the local or the other global bidder wins license B. If the local bidder B wins license B (this haens with robability P ( 2 < v 4B ) (8) 19

and we integrate this in the range to ), then the rice of license A will be the maximum of local bidder A s and the other global bidder s valuations. Equation 7 analyze the case in which the other global bidder wins license B. This haens with robability P ( 2 > v 4B ) (in the range to ), then it can enjoy synergy and will bid until v 2A + α for license A. Then, the rice global bidder 1 will ay is Max{(v 2A + α), v 3A }. Equation 8 analyzes the case in which Firm 1 wins license B, and then continue otimally on license A. This is the same as one global bidder case. The only difference is that the other global bidder is now a local bidder; hence, there is one more local bidder in the license A auction comared to the one global case. Since 2 is not known, the global bidder must use its distribution! By subtracting the second equation from the first one and equating it to zero, we will define an imlicit function of 1 and 2 as F 1 ( 1, 2) = 0 Now, we can similarly (symmetrically) define two equations for global bidder 2, and calculate F 2 ( 1, 2) = 0. To find the otimal dro out rice, one has to solve these two non-linear equations simultaneously for 1 and 2. However, the question of how to find distribution of i which is not known makes the solution imossible. We are not clear what simulation techniques may solve this roblem so we leave this as an oen roblem. 5 Conclusion and Discussion We showed the otimal bidding strategies of global bidders when there are moderate synergies and the licenses are heterogeneous. We also analyzed exosure roblem. We were able to show exosure roblem can occur even when the global bidder wins all licenses. Literature has not studies heterogeneous license case with moderate synergies since it is technically challenging when one uses more than one global bidder. With this aer, we fill this ga. One of our contributions is to write a comlicated code to calculate the 20

robability of exosure roblem. Our simulation results show that the exosure roblem may be minor for some distributions but may be u to 4.3 er cent for some others. Extending the results to n global bidders would be very comlicated since the otimal strategies of global bidders (otimal dro out rices) should be determined jointly which in turn would deend on how many local and how many global bidders are still in the auction. Moreover, one has to know the distribution of the other global bidder s otimal dro out rice while calculating the otimal dro out rice! We leave this as an oen roblem and followed the literature that use only one global bidder (e.g. Kagel and Levin (2005)). Our other contribution is comaring the revenue and the efficiency roerties of the simultaneous ascending auction with those of the VCG auction. We show that when synergy level is small (α = 0.2), the simultaneous ascending auction generates aroximately 10 er cent more revenue but allocates licenses inefficiently 8 er cent of the time. 6 Aendix Proof of Lemma 4: We will rove that there is a unique otimal dro out rice by solving EΠ 1 1 = EΠ 2 1. We have four cases. Case I: In this case, we will assume +α < 1 and +α G( A )d A +(v 1B ) < 0 imlies 1 < (which in turn imlies EΠ 1 1 > 0). First, we show that there exists a unique solution that makes equations 1 and 2 equal, and this is the otimal dro out rice 1. We define a new function, J() = EΠ 1 1 EΠ 2 1. To rove uniqueness, we will show that this function is monotonically increasing and it is negative when = v 1B (by lemma 2 cannot be less than v 1B ) and is ositive when =. Hence, there must be a unique root at the interval v 1B < <. J(, m) = 1 +α (v 1B )g( A )d A. ( A )g( A )d A +α (V 1 A )g( A )d A By using (v 1B ) 1 g( A )d A = v 1B, we can re-write it as v1a ( A )g( A )d A +α (v 1A + α A )g( A )d A (v 1B ) 21

By using integration by arts twice (and using dv = g( A )d A ), we have = ( A )G( A ) ( + α A )G( A ) +α = G( A )d( A ) G( A )d A +α G( A )d A (v 1B ) + +α G( A )d( + α A ) (v 1B ) We take artial derivative of J(, m) with resect to, we have, J(,m) = [ +α G( A )] + 1 > 0 It is ositive since the term [ +α G( A )] is negative. As the lower limit of the integral increases, the value of the exression decreases (does not increase) if the term inside is non-negative which is true since it is a cumulative distribution function. We must also show that G( A ) 0 to rove this. While one can easily see that this is correct (as increases the cumulative distribution conditional on decreases), we will give a formal roof by using Leibniz s rule when necessary. G( A ) = (m 1)f() ( A = (m 1)f()( A ( 1 = (m 1)f()( A ( 1 A = [( f(v)dv 1 f(v)dv )m 1 ] f(v)dv) m 2 ( 1 f(v)dv) m 2 f(v)dv)m 1 + (m 1)f() ( f(v)dv)m 1 [ 1 + f(v)dv) m 2 A f(v)dv ] 1 f(v)dv A f(v)dv) m 1 ( 1 f(v)dv)m [ 1 + F ( A )] < 0 ( 0 only if A = 1). f(v)dv)m 1 Thus, J(, m) is monotonically increasing function of, when v 1B <. If = v 1B, then J(v 1B ) = v 1B = +α G( A v 1B )d A < 0. G( A α)d A +α v 1B G( A v 1B )d A If =, J( ) = 0 +α G( A )d A (v 1B ) > 0, then our assumtion v1a +α G( A )d A + (v 1B ) < 0 imlies that J( = ) > 0. Hence, there is a unique root in the interval v 1B < <. Next, we show that as the number of active firms in license A auction decreases, the otimal dro out rice will increase. We will use the imlicit function theorem for this: d 1 = J( 1,m) m < 0. dm J( 1,m) 1 We have already shown that J( 1,m) 1 Since J(, m) = > 0. G( A )d A +α G( A )d A (v 1B ). 22

We take artial derivative of J(, m) with resect to m, that is, J(,m) m = +α = Since G( A ) m G( A ) d m A +α G( A ) d m A G( A ) d m A = +α ln(f ( A ))G( A )d A > 0. = ln(f ( A ))G( A ) < 0. Hence, we show that G( A ) m > 0 holds. By the imlicit function theorem, we show that the otimal dro out rice increases as the number of local firms, m, decreases. Since J(,m) > 0 and J(,m) m > 0, we have, d 1 dm = F ( 1,m) m F ( 1,m) 1 < 0 Case II: In this case, we will assume that +α < 1 and +α G( A )d A +(v 1B ) > 0 which imlies 1 >. And this condition in turn imlies that EΠ 1 1 = 0. Now let J(, m) = EΠ 1 1 EΠ 2 1 J(, m) = 0 +α (V 1 A )g( A )d A 1 +α (v 1B )g( A )d A = +α G( A )d A (v 1B ). When, we take artial derivative of J(, m) with resect to, we have, J(,m) = [ +α G( A )d A ] + 1 > 0, since G( A ) < 0. Thus, J(, m) is monotonically increasing function of, when + α. Our assumtion +α G( A )d A + (v 1B ) < 0 imlies that J( = ) < 0. If = + α, then J( + α) = 0 0 (v 1B + α) > 0. Thus, there is a unique solution, 1, in the interval (, + α). Next, we show that when the number of active firms in license A auction decreases, this otimal dro out rice will increase. We take artial derivative of J(, m) with resect to m, we have, J(,m) m = +α ln(f ( A ))G( A )d A > 0. Since J(,m) d 1 = J( 1,m) m dm J( 1,m) 1 > 0 and J(,m) < 0 m > 0, we have, Case III: In this case, we will assume that + α > 1 and 1 G( A )d A + (v 1B + α 1) < 0 which imlies 1. And this condition in turn imlies that EΠ 1 1 > 0. Now let J(, m) = EΠ 1 1 EΠ 2 1 J(, m) = ( A )g( A )d A 1 (V 1 A )g( A )d A 23

= ( A )G( A ) G( A )d( A ) ( + v 1B + α A )G( A ) 1 + 1 G( A )d( + v 1B + α A ) = G( A )d A ( + v 1B + α 1) 1 G( A )d A = ( + v 1B + α 1) 1 G( A )d A We take artial derivative of J(, m) with resect to, we have, J(,m) = [ 1 G( A )] + 1 > 0 It is ositive since the term [ 1 G( A )] is negative. And we have shown that G( A ) 0. Thus, J(, m) is monotonically increasing function of, when v 1B <. If = v 1B, then J(v 1B ) = 1 G( A v 1B )d A ( + α 1) < 0. If =, J( ) = 0 1 G( A )d A (v 1B + α 1) > 0, then our assumtion 1 G( A )d A + (v 1B + α 1) < 0 imlies that J( = ) > 0. Hence, there is a unique root in the interval v 1B < <. Next, we ski to show that as the number of active firms in license A auction decreases, the otimal dro out rice will increase, since we have done this in Case I. Case IV: In this case, we will assume that + α > 1 and 1 G( A )d A + (v 1B + α 1) > 0 which imlies 1 >. And this condition in turn imlies that EΠ 1 1 = 0. Now let J(, m) = EΠ 1 1 EΠ 2 1 J(, m) = 0 1 G( A )d A ( + v 1B + α 1). When >, we take artial derivative of J(, m) with resect to, we have, J(,m) = [ 1 G( A )d A ] + 1 > 0, since G( A ) < 0. Thus, J(, m) is monotonically increasing function of, when 1. Our assumtion 1 G( A )d A + (v 1B + α 1) > 0 imlies that J( = ) < 0. If = 1, then J(1) = 0 0 (v 1B + + α 2) > 0. Since v 1B + + α < 2. Thus, there is a unique solution, 1, in the interval (, 1). We also ski to show that when the number of active firms in license A auction decreases, this otimal dro out rice will increase which has been roven in Case II. Proof of Proosition 7: Suose that there is one global bidder and one local bidder 24

bidding for each license. In the VCG auction, the seller s revenue will be as follows deending on the valuations of the bidders. CASE I: Suose that the valuations are such that + v 1B + α < v 2A + v 3B. A) (Local bidders win each license) And suose that < v 2A and v 1B < v 3B. There are four sub cases to consider. i) v 3B > v 1B + α and v 2A > + α, then the revenue is + v 1B. since W (x 2 ) W 2 (x) = + v 3B (0 + v 3B ) = and W (x 3 ) W 3 (x) = v 1B + v 2A v 2A = v 1B ii) v 3B < v 1B + α and v 2A < + α, then the revenue is 2( + v 1B + α) v 3B v 2A. since W (x 2 ) W 2 (x) = + v 1B + α (0 + v 3B ) = + v 1B + α v 3B and W (x 3 ) W 3 (x) = + v 1B + α (0 + v 2A ) = + v 1B + α v 2A iii) v 3B > v 1B + α and v 2A < + α, then the revenue is 2 + v 1B + α v 2A. since W (x 2 ) W 2 (x) = + v 3B (0 + v 3B ) =, and W (x 3 ) W 3 (x) = + v 1B + α (0 + v 2A ) = + v 1B + α v 2A iv) v 3B < v 1B + α and v 2A > + α, then the revenue is + 2v 1B + α v 3B. since W (x 2 ) W 2 (x) = + v 1B + α (0 + v 3B ) = + v 1B + α v 3B and W (x 3 ) W 3 (x) = v 1B + v 2A v 2A = v 1B B) (Local bidder wins A, and global bidder wins B) And suose that v 2A > and v 3B < v 1B. Then, the revenue is + α + v 3B. since W (x 2 ) W 2 (x) = + v 1B + α (0 + v 1B ) = + α, and W (x 1 ) W 1 (x) = v 3B + v 2A v 2A = v 3B C) (Local bidder wins B, and global bidder wins A) And suose that v 2A < and v 3B > v 1B. Then, the revenue is v 1B + α + v 2A since W (x 3 ) W 3 (x) = + v 1B + α (0 + ) = v 1B + α, and W (x 1 ) W 1 (x) = v 3B + v 2A v 3B = v 2A CASE II: Suose that the valuations are such that + v 1B + α > v 2A + v 3B. A) (Global bidder wins both licenses) And suose that < v 2A and v 1B < v 3B. Then, the revenue is v 2A + v 3B. 25

since W (x 1 ) W 1 (x) = v 2A + v 3B (0 + 0) = v 2A + v 3B. B) (Global bidder wins both licenses) And suose that > v 2A and v 1B > v 3B. Then, the revenue is v 2A + v 3B. since W (x 1 ) W 1 (x) = v 2A + v 3B (0 + 0) = v 2A + v 3B. C) (Global bidder wins license A, local bidder B wins license B) And suose that v 2A < and v 3B > v 1B + α. Then, the revenue is v 2A + v 1B + α. since W (x 1 ) W 1 (x) = v 2A + v 3B (v 3B + 0) = v 2A and W (x 3 ) W 3 (x) = + v 1B + α ( + 0) = v 1B + α. D) (Global bidder wins license B, local bidder A wins license A) And suose that v 2A > + α and v 3B < v 1B. Then, the revenue is v 3B + + α. since W (x 1 ) W 1 (x) = v 2A + v 3B (v 2A + 0) = v 3B and W (x 2 ) W 2 (x) = + v 1B + α (v 1B + 0) = + α. 7 References 1. Albano, Gian L., Germano, Fabrizio, and Stefano Lovo (2001), A Comarison of Standard Multi-Unit Auctions with Synergies, Economics Letters, 71, 55-60. 2. Albano, Gian L., Germano, Fabrizio, and Stefano Lovo (2006), Ascending Auctions for Multile Objects: the Case for the Jaanese Design, Economic Theory, 28, 331-355. 3. Ausubel, L. M. and P. Milgrom. (2006), The Lovely But Lonely Vickrey Auction, in P. Cramton, Y. Shoham, and R Steinberg s edition, Combinatorial Auctions, MIT Press. 3. Brusco, Sandro, and Giusee Loomo (2002), Collusion via Signaling in Simultaneous Ascending Bid Auctions with Heterogeneous Objects, with and without Comlementarities, The Review of Economic Studies, 69, 407-463. 4. Brusco, Sandro, and Giusee Loomo (2009), Simultaneous Ascending Auctions with Comlementarities and Known Budget Constraints, Economic Theory, 38, 105-124. 5. Chakraborty, Indranil (2004), Multi-Unit Auctions with Synergy, Economics Bulletin, 26

4, 1-14. 6. Chow, Yuenleng and Abdullah Yavas (2009), Auctions With Positive Synergies: Exerimental Evidence, Manuscrit. 7. Englmaier, Florian, Guillen, Pablo, Llorente, Loreto, Onderstal, Sander, and Ruert Sausgruber (2009) The Chostick Auction: A Study of the Exosure Problem in Multi- Unit Auctions, International Journal of Industrial Organization, 27, 286-291. 8. Goerre, Jacob K., Lien Yuanchuan. (2010) An Equilibrium Analysis of the Simultaneous Ascending Auction, Working aer. 9. Harstad, R and M. Rothkof (1995) Withdrawable Bids as Winner s Curse Insurance, Oerations Research, 43, 983-994 10. Kagel, John H. and Dan Levin (2005) Multi-Unit Demand Auctions with Synergies: Behavior in Sealed-Bid versus Ascending-Bid Uniform-Price Auctions, Games and Economic Behavior, 53, 170-207. 11. Katok, Elena and Alvin E. Roth (2004) Auctions of Homogeneous Goods with Increasing Returns: Exerimental Comarison of Alternative Dutch Auctions, Management Science, 50, 1044-1063. 12. Krishna, Vijay and Robert W. Rosenthal (1996), Simultaneous Auctions with Synergies, Games and Economic Behavior, 17, 1-31. 13. Porter, David P.(1999), The Effect of Bid Withdrawal in a Multi-Object Auction, Review of Economic Design, 4, 73-97. 14. Menicucci, Domenico, (2003), Otimal Two-Object Auctions with Synergies, Review of Economic Design, 8, 143-164. 15. Rosenthal, Robert W. and Ruqu Wang, (1996), Simultaneous Auctions with Synergies and Common Values, Games and Economic Behavior, 17, 32-55. 16. Sano, Ryuji, (2011), Non-bidding equilibrium in an ascending core-selecting auction, 27

Games and Economic Behavior, forthcoming. doi:10.1016/j.geb.2011.08.016 17. Wang, R. (2000), Bidding and Renegotiation in Procurement Auctions, Euroean Economic Review, 44, 1577-1597. 18. Zheng, C. Z. (2008) Jum Bidding and Overconcentration in Decentralized Simultaneous Ascending Auctions, Working Paer. 28